0038 — ARIA Learning Model: Observation Log + SQL Aggregates (Trade DNA Genetic Algorithm Retired)

Status

Folded into aria-companion.md — defer to that page for the canonical description. (re-verified 2026-08-06)

Context

The original ARIA learning surface was framed as a per-player genetic algorithm — every ARIATradingPattern row was a "DNA" genome with mutation, generation lineage, fitness scoring, and a dedicated anti-gaming layer defending the fitness-gaming surface that framing created. This ADR retired that model in favor of an immutable per-trade observation log (ARIATradingObservation) mined by explicit SQL aggregates and heuristics — no genome, no fitness, no anti-gaming layer required, and every recommendation carries a plain-English explanation sourced from the aggregate that produced it. The privacy-first per-player isolation rule from ADR-0016 is unaffected — this ADR changes only the learning mechanism, not its scope.

Folded into: FEATURES/gameplay/aria-companion.md (player-facing "How ARIA learns" section), OPERATIONS/aria.md (ARIATradingObservation schema + recommendation-engine spec).

Related docs

• OPERATIONS/aria.md
• FEATURES/gameplay/aria-companion.md
• DATA_MODELS/player.md
• DATA_MODELS/jsonb-schema.md
• ADR/0016-a1-per-player-aria-no-aggregate-ml.md — per-player isolation rule this ADR honors and does not change.

Addendum — 2026-07-10 (Max ruling: table drop approved)

Per ADR-0093 item 26, DROP TABLE aria_trading_patterns is approved — completing the removal this ADR already specified at the ORM-relationship level.